Improving the accuracy of two-sample summary-data Mendelian randomization: moving beyond the NOME assumption.

Bowden J, Del Greco M F, Minelli C, Zhao Q, Lawlor DA, Sheehan NA, Thompson J, Davey Smith G. · 2019

Read Improving the accuracy of two-sample summary-data Mendelian randomization: moving beyond the NOME assumption. by Bowden J, Del Greco M F, Minelli C, Zhao Q, Lawlor DA, Sheehan NA, Thompson J, Davey Smith G. (2019) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc6659376.txt.

Introduction

Bowden and co-authors improve the weighting used in two-sample Mendelian randomization. Their methods paper appears in the International Journal of Epidemiology. The technique pools causal estimates from many genetic variants. Each variant's ratio estimate needs an inverse-variance weight. Standard first-order weights can fake heterogeneity that is not there. Second-order weights can hide heterogeneity that is real. Both flaws trace to the NOME assumption. The team derives modified weights to fix them.

Monte Carlo simulations favoured the new weights. They quantified heterogeneity more accurately than either rival. They also removed regression dilution bias from weak instruments. That gain costs some precision with few weak instruments. An example revisits blood pressure and heart disease. The authors recommend pairing the schemes in practice.

Catalog record

CREATORBowden J, Del Greco M F, Minelli C, Zhao Q, Lawlor DA, Sheehan NA, Thompson J, Davey Smith G.
YEAR2019
COLLECTIONPAPERS
FORMATTXT
FILE SIZE46 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC6659376

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Improving the accuracy of two-sample summary-data Mendelian randomization: moving beyond the NOME assumption By Jack Bowden, Fabiola Del Greco M, Cosetta Minelli, Qingyuan Zhao, Debbie A Lawlor, Nuala A Sheehan, John Thompson, George Davey Smith Published in: International Journal of Epidemiology DOI: 10.1093/ije/dyy258 ABSTRACT Abstract Background Two-sample summary-data Mendelian randomization (MR) incorporating multiple genetic variants within a meta-analysis framework is a popular technique for assessing causality in epidemiology. If all genetic variants satisfy the instrumental variable (IV) and necessary modelling assumptions, then their individual ratio estimates of causal effect should be homogeneous. Observed heterogeneity signals that one or more of these assumptions could have been violated. Methods Causal estimation and heterogeneity assessment in MR require an approximation for the variance, or equivalently the inverse-variance weight, of each ratio estimate. We show that the most popular ‘first-order’ weights can lead to an inflation in the chances of detecting heterogeneity when in fact it is not present. Conversely, ostensibly more accurate ‘second-order’ weights can dramatically increase the chances of failing to detect heterogeneity when it is truly present. We derive modified weights to mitigate both of these adverse effects. Results Using Monte Carlo simulations, we show that the modified weights outperform first- and second-order weights in terms of heterogeneity quantification. Modified weights are also shown to remove the phenomenon of regression dilution bias in MR estimates obtained from weak instruments, unlike those obtained using first- and second-order weights. However, with small numbers of weak instruments, this comes at the cost of a reduction in estimate precision and power to detect a causal effect compared with first-order weighting. Moreover, first-order weights always furnish unbiased estimates and preserve the type I error rate under the causal null. We illustrate the utility of the new method using data from a recent two-sample summary-data MR analysis to assess the causal role of systolic blood pressure on coronary heart disease risk. Conclusions We propose the use of modified weights within two-sample summary-data MR studies for accurately quantifying heterogeneity and detecting outliers in the presence of weak instruments. Modified weights also have an important role to play in terms of causal estimation (in tandem with first-order weights) but further research is required to understand their strengths and weaknesses in specific settings. Keywords: Two-sample summary-data Mendelian randomization, inverse-variance weighted estimate, Cochran’s Q statistic, outlier detection Key Messages Two-sample summary-data Mendelian randomization requires the specification of inverse-variance weights for model fitting, heterogeneity quantification and outlier detection amongst a set of causal estimates.Heterogeneity indicates a possible violation of the necessary IV or modelling assumptions of which pleiotropy is a likely major cause.First-order weights can inflate the type I error rate of Cochran’s Q statistic for detecting heterogeneity about the inverse-variance weighted (IVW) estimate when the NO Measurement Error (NOME) assumption is strongly violated (as judged by a low F-statistic) and the true causal effect of interest is non-zero.Second-order weights can reduce the power of Cochran’s Q statistic for detecting heterogeneity about the IVW estimate when the NOME assumption is violated.Modified weights (developed in this paper) preserve the type I error rate of Cochran’s Q statistic, whilst maintaining its statistical power.‘Exact’ modified weights should be used for global tests of heterogeneity. ‘Iterative’ modified weights should be used to assess the outlier status of individual single-nucleotide polymorphisms (SNPs).IVW estimates obtained using exact weights are naturally corrected for regression dilution bias, and work well with large numbers of instruments, but can be imprecise relative to other weighting schemes with small numbers of weak instruments.Regardless of the number or strength of instruments used, first-order weights always furnish unbiased IVW estimates and preserve the type I error rate under the causal null. INTRODUCTION Mendelian randomization (MR)1 is an instrumental variable approach that uses genetic data, typically in the form of single-nucleotide polymorphisms (SNPs), to assess whether a modifiable exposure exerts a causal effect on a health outcome in the presence of unmeasured confounding. A particular MR study design gaining in popularity instead combines publically available summary data on SNP–exposure and SNP–outcome associations from two separate studies for large numbers of uncorrelated variants within the framework of a meta-analysis. These studies should contain no overlapping individuals (to ensure independence) but should also originate from the same source population. This is referred to as two-sample summary-data MR.2 Providing the necessary modelling assumptions are met and the chosen set of SNPs are all valid instrumental variables, an inverse-variance weighted (IVW) average of their individual causal ratio estimates provides an efficient and consistent estimate for the causal effect. This is referred to as the IVW estimate (see Box 1). Cochran’s Q statistic, which is derived from the IVW estimate, should follow a χ2 distribution with degrees of freedom equal to the number of SNPs minus 1. Excessive heterogeneity is an indication that either the modelling assumptions have been violated, or that some of the genetic variants violate the IV assumptions—e.g. by exerting a direct effect on the outcome not through the exposure.3 This is termed ’horizontal pleiotropy’.4,5 For brevity, we will refer to problematic horizontal pleiotropy simply as pleiotropy from now on. Box 1. Standard two-sample summary-data MR The IV assumptions: The canonical approach to MR assumes that a group of SNPs are valid IVs for the purposes of inferring the causal effect of an exposure, X, on an outcome, Y. That is, they are: associated with X (IV1); not associated with any confounders of X and Y (IV2); and can only be associated with Y through X (IV3). The IV assumptions are represented by the solid lines in the causal diagram below for a SNP Gj, with unobserved confounding represented by U. Dotted lines represent dependencies between G and U, and G and Y that are prohibited by the IV assumptions. The causal effect of a unit increase in X on the outcome Y, denoted by β, is the quantity we are aiming to estimate. The ratio estimate: Assume that exposure X causally affects outcome Y linearly across all values of X, so that a hypothetical intervention that induced a 1-unit increase in X would induce a β increase in Y. Suppose also that all L SNPs predict the exposure via an additive linear model with no interactions. If SNP j is a valid IV, and the two study samples are homogeneous, then the underlying SNP–outcome association from Sample 1, Γj, should be a scalar multiple of the underlying SNP–exposure association estimate from Sample 2, γj, the scalar multiple being the causal effect β. That is: Γj=βγj.The ratio estimate for the causal effect of X on Y using SNP j (out of L), β^j=Γ^j/γ^j, where Γ^j is the estimate for SNP j’s association with the outcome (with standard error σY j) and γ^j is the estimate for SNP j’s association with the exposure (with standard error σXj). The IVW estimate: The overall inverse-variance weighted (IVW) estimate for the causal effect obtained across L uncorrelated SNPs is then given by β^IVW=∑j=1Lwjβ^j∑j=1Lwj,where wj is the inverse-variance of β^j. Cochran’s Q statistic: (1)Q=∑j=1LQj=∑j=1Lwj(β^j−β^IVW)2,can then be used to test for the presence of heterogeneity. If heterogeneity is detected, this provides evidence of horizontal pleiotropy. Two popular choices for the inverse-variance weights used to calculate the IVW estimate and Cochran’s Q statistic are: 1st

Read the complete text